Is the Post-it video fair and balanced? — a claim-check with live elections¶

The lesson page for the whole Post-it set: a claim-by-claim check of Equal Vote's video "Updated: How does RCV work? — With Post-its!", with every claim reproduced (or stress-tested) as a live BetterVoting election and an LH tabulation you can re-run.

The three live elections:

BV2176 p8dp28 — the video's election, three races (STAR / RCV-IRV / Ranked Robin) → case page
BV2177 v8r66y — the same 20 voters, the single-winner methods side by side (STAR, Ranked Robin, Approval, RCV-IRV, Choose-One), four different winners → case page
BV2178 8kg698 — the video's round-2 hypothetical made real by a two-ballot flip → case page

What the video gets right¶

The RCV-IRV count is exactly correct. Every round reproduces digit-for-digit in two independent tabulators (the LH engine and BetterVoting's live count): 7/6/4/3 → Pink out → 8/7/4 → Blue out → Purple 9, Green 8, with 3 of 20 ballots exhausted. No quibble anywhere.

The core criticism is real, and it needs no scores to state. RCV-IRV eliminated Blue in round 2 without ever asking the head-to-head question, and on these very ballots Blue beats Purple 10–9. That fact is pure rank arithmetic — count, for each ballot, which of the two is ranked higher. The winner was decided by which candidate got eliminated, not by who voters preferred between the finalists-that-might-have-been.

The exhausted-ballot and shrinking-majority points are fair. Purple's final "majority" is 9 of 17 still-active ballots (53%) — but 9 of 20 cast (45%). Three voters' ballots ran out of rankings and simply left the room before the final question was asked.

Where a sharp critic pushes back¶

1. The "what if Green were eliminated instead?" scenario is a hypothetical IRV would never execute. At that point Green held 7 votes to Blue's 4; last place is eliminated, full stop. Stated as an elimination counterfactual, it can sound like IRV mis-executed its own rules — it didn't. The honest version of the point is the pairwise fact above (Blue beats Purple 10–9), which needs no counterfactual. And it can be made real: in BV2178, two of the six Green>Blue>Pink voters genuinely prefer Blue first — and RCV-IRV then lands the video's exact hypothetical tally, Blue 10, Purple 9. Two ballots is all the elimination order's cliff-edge needs.

2. There is no "rightful winner" here to be robbed of. The video's framing suggests RCV-IRV denied the stronger candidate. But this electorate is a Condorcet cycle: Blue beats Purple 10–9, yes — but Green beats Blue 7–4, and Purple beats Green 9–8 (and Pink beats Purple 12–8). Every candidate loses to somebody. Blue is not a Condorcet winner the way, say, Nashville is in the Tennessee example; Blue merely wins the particular matchup STAR's runoff happens to hold. BetterVoting's own Ranked Robin race elects Green on these ballots (the LH engine's ladder says Blue — a documented tiebreak divergence, not a bug). A fully balanced telling is harsher on IRV and humbler for STAR: what's damning for IRV is not that it missed "the right winner" — none exists — it's that it never looked at the evidence. What's humbling for STAR is that its winner here is a defensible choice, not the provable one.

3. The 0–5 scores are the video's invention — and the gap sizes do the work. The 20 voters never actually scored anyone; the video assigned scores to its own ranked ballots. To the video's credit, the assignment is clean: all 20 score ballots are strictly order-consistent with their rankings (we verified this programmatically, ballot by ballot — every ranked candidate outscores every unranked one, and every rank step is a strict score drop). So the scores are correct as an expression of the ranks. But they are not implied by the ranks — and that difference decides the election:

Scoring of the same rankings	Scoring round (P/G/B/K)	Finalists	STAR winner
The video's generous gaps (5,4,3 down-ballot)	46 / 38 / 44 / 44	Purple & Blue	Blue 10–9
Stingy gaps (5,2,1 down-ballot)	40 / 34 / 32 / 26	Purple & Green	Purple 9–8

Same 20 rankings, both scorings strictly consistent with them — and STAR flips between disagreeing with IRV (Blue) and agreeing with it (Purple). Blue's spot in the runoff hangs on the six Green fans scoring their second choice a generous 4 (and even then Blue only ties Pink 44–44 for the slot, advancing on the head-to-head). This is not a gotcha — it is STAR's entire premise: preference strength is real information that ranked ballots cannot carry, and STAR's answer legitimately depends on it. A video demonstrating STAR must invent that information; a real STAR election collects it from the voters. The fair conclusion is narrower than the video implies: if voters feel the way the video's scores say, STAR elects Blue — and IRV elects Purple either way, because it never asks.

4. Conversions are never neutral — the Approval race proves it in one line. The video's scores use only 0/3/4/5, so converting them to Approval ballots at any threshold from 1 to 3 gives identical votes — and elects Pink (12 approvals, riding the Green fans' 3-for-their-last-choice). Approve only 4s and 5s: Blue wins. Only 5s: Purple wins. Three defensible thresholds, three winners (live as BV2177's Approval race, threshold ≤ 3). Anyone who says "just simplify to Approval" has chosen a winner and may not know it.

The verdict¶

Fair on the count, advocacy in the frame. The video's arithmetic is flawless and its central complaint — IRV decides by elimination order and never consults the head-to-head — is true and important. Its two rhetorical shortcuts are presenting an elimination IRV cannot perform as a near-miss (fixed live by BV2178), and presenting invented preference strengths as if they were data (stress-tested above — under stingier but equally consistent scores, STAR agrees with IRV). And it stops one twist short of full honesty about its own example: there is no consensus winner in this electorate at all — run the single-winner methods side by side on BetterVoting and all four candidates win somewhere (BV2177):

Purple	Green	Blue	Pink
RCV-IRV, Choose-One	Ranked Robin (BV's ladder)	STAR (and LH's RR ladder)	Approval (approve = any support)

Lessons to take away¶

"RCV" names a ballot, not a result. The identical ranked ballots elect Purple (IRV/STV), Green (BV's Ranked Robin), or Blue (LH's Ranked Robin ladder). Say which tabulation you mean.
IRV's real flaw here is ignorance, not error. It executed its rules perfectly — and its rules never look at 10–9. The strongest criticism is the evidence it doesn't consult, not a winner it "got wrong."
Preference strength is information. STAR's divergence from IRV on this profile exists because the scores say the Green fans strongly back Blue. Change the strength, keep the ranks, and the divergence vanishes. That is a feature when voters supply the scores — and an assumption when a video does.
Every conversion picks a winner. Ranks→scores (5,4,3 vs 5,2,1: Blue vs Purple) and scores→approvals (threshold 3/4/5: Pink/Blue/Purple) are editorial acts. Show your conversion rule or your comparison isn't reproducible.
Cycles are knife-edges. Two ballots (BV2178) dissolve the cycle into a clean Condorcet winner and snap every ballot-reading method into unanimity — while Choose-One never notices. When methods disagree, suspect a cycle; when a cycle, expect defensible disagreement (even between two correct engines).

Reproduce everything¶

Claim	Where to check
The video's IRV rounds	BV2176 IRV race · LH mirror
Blue beats Purple 10–9	any race's pairwise matrix, e.g. the RR mirror
STAR → Blue under the video's scores	BV2176/77 STAR races · LH mirror
STAR → Purple under stingy scores	re-score page · bv2176_p8dp28_star.yaml with 5→5, 4→2, 3→1 and re-run (40/34/32/26; runoff Purple 9–8)
Approval threshold picks the winner	BV2177 Approval race (threshold ≤ 3 → Pink) · LH mirror
The switch, real	BV2178 · LH mirror

file: postit_video_fair_and_balanced.md¶